Below I present the distribution for each of the individual police evaluation items. I include both the full sample, and then the white and black only samples respectively. Higher values denote more positive evaluations. Everything is in percentage points. I also report the results from a Chi\(^2\) test on these distributions. Unsurprisingly, all of these are significant.
Of these individual items, blacks tend to offer the most negative ratings on the equal treatment, excessive force, and accountability items. For whites, the distribution of repsonses to these items does not appear to meaningfully differ from the rest, at least eyeballing the results.
Solving Crime
## black
## p.crim.solve 0 1
## 0 5 19
## 1 11 20
## 2 35 36
## 3 36 17
## 4 13 7
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.crim.solve + black, d.all), 2) * 100)
## X-squared = 19.4, df = 4, p-value = 0.0006556
Protecting people like you from violent crime
## black
## p.viol.crim 0 1
## 0 4 21
## 1 9 19
## 2 28 34
## 3 40 17
## 4 20 8
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.viol.crim + black, d.all), 2) * 100)
## X-squared = 30.119, df = 4, p-value = 4.63e-06
Treating racial and ethnic groups equally
## black
## p.race.fair 0 1
## 0 12 43
## 1 13 18
## 2 30 23
## 3 29 11
## 4 16 6
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.race.fair + black, d.all), 2) * 100)
## X-squared = 31.845, df = 4, p-value = 2.058e-06
Not using excessive force on suspects
## black
## p.exces.force 0 1
## 0 9 35
## 1 13 18
## 2 31 28
## 3 31 12
## 4 16 7
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.exces.force + black, d.all), 2) * 100)
## X-squared = 28.24, df = 4, p-value = 1.115e-05
Holding police officers accountable for misconduct
## black
## p.account 0 1
## 0 12 44
## 1 12 15
## 2 29 24
## 3 32 11
## 4 15 6
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.account + black, d.all), 2) * 100)
## X-squared = 33.204, df = 4, p-value = 1.085e-06
I also considered a summary evaluation index. I summed together the 5 evaluations and set the scale to run from 0-1, where higher values denote more positive evaluations. The mean for the full sample is 0.53, while for whites it is 0.59 and for blacks it is 0.37. Blacks clearly rate the police on average lower than whites, and this difference is significant at p < 0.000. I present the distribution for the scale for the full sample and by race below.
It’s also potentially instructive to contrast whites and blacks in how these police evaluation items scale together. To get a sense for whether these capture summary evaluations across groups, I present alphas for the 5 items scaled together. Cronbach’s alpha for whites is 0.90, while for black it is 0.89. Although a rough pass, the similarity suggests that blacks and whites use the same dimensions to evaluate the police. I could push further on this with some factor analyses if interested.
Race seems closely related to how people evaluate the police in their area. This manifests both in individual and summary item ratings. Importantly, the dimensions on which whites and blacks evaluate the police seem to matter the same.
I created two separate measure of class based on tercile breakdowns of income and education. Each assigned repondents to an income or education tercile, however one version determined terciles based on the full weighted sample while the second looked within each racial group. Because the correlation between the two measures is 0.92 I use the class measure that’s specific within each race to account for potential incomparabilities across groups. I again included a Chi\(^2\) test for each distribution. None of these are significant. Class level does not appear to be related with evaluations of the police. Moreover, response distributions appear to be similar across items, too.
Solving Crime
## class
## p.crim.solve 0 0.25 0.5 0.75 1
## 0 13 10 8 6 6
## 1 16 14 14 12 11
## 2 36 36 34 35 35
## 3 24 29 33 36 34
## 4 11 10 11 10 15
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.crim.solve + class, d.all), 2) * 100)
## X-squared = 9.6774, df = 16, p-value = 0.8829
Protecting people like you from violent crime
## class
## p.viol.crim 0 0.25 0.5 0.75 1
## 0 12 10 8 6 5
## 1 16 13 11 9 11
## 2 32 32 29 29 25
## 3 25 31 35 40 38
## 4 15 15 17 16 21
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.viol.crim + class, d.all), 2) * 100)
## X-squared = 13.171, df = 16, p-value = 0.6602
Treating racial and ethnic groups equally
## class
## p.race.fair 0 0.25 0.5 0.75 1
## 0 22 22 21 17 19
## 1 16 14 14 14 13
## 2 30 29 27 29 25
## 3 19 23 26 27 28
## 4 13 12 13 13 15
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.race.fair + class, d.all), 2) * 100)
## X-squared = 4.3657, df = 16, p-value = 0.9981
Not using excessive force on suspects
## class
## p.exces.force 0 0.25 0.5 0.75 1
## 0 19 18 16 14 12
## 1 15 15 14 14 15
## 2 33 32 28 30 26
## 3 20 23 29 28 29
## 4 12 12 13 14 17
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.exces.force + class, d.all), 2) * 100)
## X-squared = 7.124, df = 16, p-value = 0.9708
Holding police officers accountable for misconduct
## class
## p.account 0 0.25 0.5 0.75 1
## 0 25 22 21 18 17
## 1 13 14 12 12 12
## 2 29 29 27 28 24
## 3 21 23 28 29 31
## 4 12 11 12 13 15
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.account + class, d.all), 2) * 100)
## X-squared = 6.305, df = 16, p-value = 0.9845
As for the summary evaluation index, the table below proveis the mean for each class category. Descriptively higher class individuals tend to evaluate the police more positively. A 5 point difference exists between the lowest and highest class individuals, one significant at p < 0.000.
## 0 0.25 0.5 0.75 1
## mean 0.51 0.52 0.53 0.54 0.56
The plots below present the distribution of summary police evaluations for each class level.
I return to the alpha measure to contrast class category groups’ police evaluations. The table below presents these tallies. No meaningful variation exists by class category, suggesting class does not shape which dimensions people rely on for evaluating the police.
## 0 0.25 0.5 0.75 1
## alpha 0.915 0.907 0.914 0.914 0.919
To summarize, class appears unrelated to individuals’ evaluations of the police. This holds for both the individual items and the summary index.
Finally, for the race and class breakdown I present the item distributions again, but by class within each racial group. I again include Chi\(^2\) tests to compare the distributions. None of these tests are significant, suggesting that the intersection of race and class does not affect evaluations of the police.
Whites: Solving Crime
## class
## p.crim.solve 0 0.25 0.5 0.75 1
## 0 7 6 4 5 4
## 1 14 12 12 10 10
## 2 37 36 34 35 34
## 3 30 33 38 38 36
## 4 13 12 12 12 16
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.crim.solve + class, d.wht), 2) * 100)
## X-squared = 4.7253, df = 16, p-value = 0.997
Blacks: Solving Crime
## class
## p.crim.solve 0 0.25 0.5 0.75 1
## 0 20 19 18 12 12
## 1 22 22 21 21 13
## 2 36 35 36 39 44
## 3 14 17 19 24 24
## 4 8 7 6 3 7
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.crim.solve + class, d.blk), 2) * 100)
## X-squared = 14.471, df = 16, p-value = 0.5636
Whites: Protecting people like you from violent crime
## class
## p.viol.crim 0 0.25 0.5 0.75 1
## 0 6 5 3 4 1
## 1 13 11 9 7 8
## 2 31 32 29 26 23
## 3 32 35 40 44 43
## 4 19 18 19 19 24
## Warning in chisq.test(round(prop.table(svytable(~p.viol.crim + class,
## d.wht), : Chi-squared approximation may be incorrect
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.viol.crim + class, d.wht), 2) * 100)
## X-squared = 12.146, df = 16, p-value = 0.7338
Blacks: Protecting people like you from violent crime
## class
## p.viol.crim 0 0.25 0.5 0.75 1
## 0 20 21 20 16 21
## 1 22 19 17 18 20
## 2 36 33 34 39 31
## 3 12 21 20 22 19
## 4 10 7 9 4 10
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.viol.crim + class, d.blk), 2) * 100)
## X-squared = 9.3279, df = 16, p-value = 0.8993
Whites: Treating racial and ethnic groups equally
## class
## p.race.fair 0 0.25 0.5 0.75 1
## 0 13 14 13 11 11
## 1 16 13 12 13 13
## 2 31 31 30 30 26
## 3 24 27 30 30 32
## 4 16 15 15 16 18
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.race.fair + class, d.wht), 2) * 100)
## X-squared = 3.5939, df = 16, p-value = 0.9994
Blacks: Treating racial and ethnic groups equally
## class
## p.race.fair 0 0.25 0.5 0.75 1
## 0 37 40 44 40 51
## 1 18 17 20 24 15
## 2 28 25 21 23 18
## 3 10 11 12 12 10
## 4 7 6 4 2 6
## Warning in chisq.test(round(prop.table(svytable(~p.race.fair + class,
## d.blk), : Chi-squared approximation may be incorrect
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.race.fair + class, d.blk), 2) * 100)
## X-squared = 11.3, df = 16, p-value = 0.7906
Whites: Not using excessive force on suspects
## class
## p.exces.force 0 0.25 0.5 0.75 1
## 0 12 11 10 9 7
## 1 14 14 12 14 14
## 2 34 34 29 30 27
## 3 26 27 33 32 33
## 4 15 15 16 16 19
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.exces.force + class, d.wht), 2) * 100)
## X-squared = 5.2147, df = 16, p-value = 0.9946
Blacks: Not using excessive force on suspects
## class
## p.exces.force 0 0.25 0.5 0.75 1
## 0 31 35 35 34 32
## 1 17 18 19 19 21
## 2 34 29 26 30 22
## 3 11 13 14 11 15
## 4 6 5 5 5 9
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.exces.force + class, d.blk), 2) * 100)
## X-squared = 6.7311, df = 16, p-value = 0.9781
Whites: Holding police officers accountable for misconduct
## class
## p.account 0 0.25 0.5 0.75 1
## 0 16 14 12 12 10
## 1 13 14 11 12 12
## 2 30 31 30 28 25
## 3 26 27 33 34 36
## 4 14 13 14 15 17
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.account + class, d.wht), 2) * 100)
## X-squared = 5.9786, df = 16, p-value = 0.9883
Blacks: Holding police officers accountable for misconduct
## class
## p.account 0 0.25 0.5 0.75 1
## 0 40 41 47 39 50
## 1 15 16 16 17 13
## 2 28 26 20 29 18
## 3 12 12 12 11 10
## 4 6 6 5 4 9
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~p.account + class, d.blk), 2) * 100)
## X-squared = 9.3722, df = 16, p-value = 0.8973
Returning to the summary evaluation index, the table below provides the means for each race/class category. Whereas the prior class-only results indicated that higher class individuals tended to evaluate the police more positively, this seems driven by whites. A 7 point difference exists between the lowest and highest class whites, but this gap is only 2 points for blacks. The former is significant at p < 0.000 while the latter is not (p = 0.703).
## 0 0.25 0.5 0.75 1
## mean - White 0.56 0.57 0.60 0.60 0.63
## mean - Black 0.35 0.38 0.35 0.37 0.37
Finally, I present the scale alphas in table below. The first row looks at whites across class, while the second looks at blacks by class. No meaningful variation exists according to class/race interaction, reinforcing the likelihood that people rely on the same dimensions for evaluating the police.
## 0 0.25 0.5 0.75 1
## Alpha - Whites 0.906 0.901 0.897 0.903 0.898
## Alpha - Blacks 0.900 0.884 0.898 0.891 0.907
I break down each court fairness item based on the suffix. The first is whether the court will fairly apply the law, while the second two ask whether this is the case regardless of a person’s class or race, resepctively. Again, I presented the response distribution in percentage points, broken down by race. The Chi\(^2\) tests are again significant. Regardless of the prompt, blacks are on average less likely to think the courts in their area will be fair.
‘’fairly apply the law?’’
## black
## court.fair 0 1
## 0 5 16
## 0.333333333333333 14 29
## 0.666666666666667 52 43
## 1 29 12
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~court.fair + black, d.all), 2) * 100)
## X-squared = 18.896, df = 3, p-value = 0.0002873
‘’fairly apply the law, regardless of a person’s class?’’
## black
## court.fair.class 0 1
## 0 6 16
## 0.333333333333333 15 27
## 0.666666666666667 49 43
## 1 30 14
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~court.fair.class + black, d.all), 2) * 100)
## X-squared = 14.184, df = 3, p-value = 0.002666
‘’fairly apply the law, regardless of a person’s race?’’
## black
## court.fair.race 0 1
## 0 6 21
## 0.333333333333333 14 30
## 0.666666666666667 45 37
## 1 35 12
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~court.fair.race + black, d.all), 2) * 100)
## X-squared = 26.187, df = 3, p-value = 8.714e-06
We also see interesting treatment effects within racial group. While there are no differences between the baseline condition and the class prime, the race prime decreases blacks’ perceptions that courts will be fair. In contrast, the same prime increases whites’ perceptions of fairness. These differences are small, however. The Cohen’s D effect size for whites is 0.06, whole for blacks it is 0.12. Even so, because of the divergent effects, the black-white gap in fairness evaluations grows by 5 percentage points, from 18 to 23 points.
##
## Call:
## lm(formula = court.fair.all ~ court.fair.treat * black, data = cjs.df,
## weights = wts_whole)
##
## Weighted Residuals:
## Min 1Q Median 3Q Max
## -1.70289 -0.13145 -0.01167 0.21515 1.30115
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.681538 0.005430 125.504 < 2e-16 ***
## court.fair.treatRace 0.013662 0.007723 1.769 0.076919 .
## court.fair.treatClass -0.004915 0.007688 -0.639 0.522699
## black -0.177522 0.010409 -17.055 < 2e-16 ***
## court.fair.treatRace:black -0.048872 0.014807 -3.301 0.000967 ***
## court.fair.treatClass:black 0.020124 0.014756 1.364 0.172677
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.2839 on 11156 degrees of freedom
## (4 observations deleted due to missingness)
## Multiple R-squared: 0.08072, Adjusted R-squared: 0.0803
## F-statistic: 195.9 on 5 and 11156 DF, p-value: < 2.2e-16
Turning to class, the analyses below suggest little variation exists by class category in fairness percpetions. Moreover, this holds regardless of the prompt. Even when primed to think about class, low and high class respondents think the courts in their area will fairly apply the law. ``fairly apply the law?’’
## class.rac
## court.fair 0 0.25 0.5 0.75 1
## 0 11 7 9 6 6
## 0.333333333333333 20 20 17 20 15
## 0.666666666666667 47 50 51 48 53
## 1 22 23 24 27 26
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~court.fair + class.rac, d.all), 2) * 100)
## X-squared = 4.72, df = 12, p-value = 0.9667
``fairly apply the law, regardless of a person’s class?’’
## class.rac
## court.fair.class 0 0.25 0.5 0.75 1
## 0 10 9 9 9 6
## 0.333333333333333 19 19 17 19 16
## 0.666666666666667 49 46 49 45 48
## 1 22 25 24 27 31
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~court.fair.class + class.rac, d.all), 2) * 100)
## X-squared = 3.555, df = 12, p-value = 0.9902
``fairly apply the law, regardless of a person’s race?’’
## class.rac
## court.fair.race 0 0.25 0.5 0.75 1
## 0 12 11 9 8 9
## 0.333333333333333 17 19 20 19 14
## 0.666666666666667 41 44 40 43 48
## 1 30 25 30 30 29
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~court.fair.race + class.rac, d.all), 2) * 100)
## X-squared = 3.9198, df = 12, p-value = 0.9848
##
## Call:
## lm(formula = court.fair.all ~ court.fair.treat * class.rac, data = cjs.df,
## weights = wts_whole)
##
## Weighted Residuals:
## Min 1Q Median 3Q Max
## -1.62974 -0.21365 0.02581 0.07588 0.91059
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 6.072e-01 8.221e-03 73.856 < 2e-16 ***
## court.fair.treatRace 7.624e-03 1.176e-02 0.648 0.517
## court.fair.treatClass 3.421e-05 1.164e-02 0.003 0.998
## class.rac 5.662e-02 1.447e-02 3.913 9.19e-05 ***
## court.fair.treatRace:class.rac -1.513e-02 2.089e-02 -0.724 0.469
## court.fair.treatClass:class.rac 1.509e-03 2.046e-02 0.074 0.941
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.2956 on 11156 degrees of freedom
## (4 observations deleted due to missingness)
## Multiple R-squared: 0.00349, Adjusted R-squared: 0.003043
## F-statistic: 7.814 on 5 and 11156 DF, p-value: 2.37e-07
Finally, looking at the intersection of race and class, little variation again appears by class level. One interesting point is that for blacks, the class prime appears to have decreases the number of lower class blacks believing the courts in their area will fairly apply the law. The p-vale on the Chi\(^2\) test is 0.082.
Whites: ``fairly apply the law?’’
## class.rac
## court.fair 0 0.25 0.5 0.75 1
## 0 8 5 5 4 4
## 0.333333333333333 19 15 12 14 14
## 0.666666666666667 48 52 55 50 52
## 1 26 27 28 32 29
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~court.fair + class.rac, d.wht), 2) * 100)
## X-squared = 5.0985, df = 12, p-value = 0.9546
Blacks: ``fairly apply the law?’’
## class.rac
## court.fair 0 0.25 0.5 0.75 1
## 0 22 11 16 11 12
## 0.333333333333333 26 35 27 35 20
## 0.666666666666667 43 43 44 46 53
## 1 9 12 13 8 15
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~court.fair + class.rac, d.blk), 2) * 100)
## X-squared = 16.43, df = 12, p-value = 0.1723
Whites: ``fairly apply the law, regardless of a person’s class?’’
## class.rac
## court.fair.class 0 0.25 0.5 0.75 1
## 0 6 8 7 7 2
## 0.333333333333333 20 18 13 13 10
## 0.666666666666667 50 46 50 48 51
## 1 24 27 30 32 37
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~court.fair.class + class.rac, d.wht), 2) * 100)
## X-squared = 11.76, df = 12, p-value = 0.4651
Blacks: ``fairly apply the law, regardless of a person’s class?’’
## class.rac
## court.fair.class 0 0.25 0.5 0.75 1
## 0 25 12 16 15 15
## 0.333333333333333 16 23 29 34 29
## 0.666666666666667 46 45 47 38 41
## 1 13 20 9 12 15
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~court.fair.class + class.rac, d.blk), 2) * 100)
## X-squared = 19.288, df = 12, p-value = 0.08181
Whites: ``fairly apply the law, regardless of a person’s race?’’
## class.rac
## court.fair.race 0 0.25 0.5 0.75 1
## 0 8 8 4 6 5
## 0.333333333333333 14 17 16 13 10
## 0.666666666666667 44 44 44 44 52
## 1 34 30 36 37 33
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~court.fair.race + class.rac, d.wht), 2) * 100)
## X-squared = 6.2091, df = 12, p-value = 0.9052
Blacks: ``fairly apply the law, regardless of a person’s race?’’
## class.rac
## court.fair.race 0 0.25 0.5 0.75 1
## 0 28 20 19 14 21
## 0.333333333333333 25 24 32 36 32
## 0.666666666666667 33 44 36 45 35
## 1 14 12 14 5 11
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~court.fair.race + class.rac, d.blk), 2) * 100)
## X-squared = 16.484, df = 12, p-value = 0.17
However, we get more nuance by looking at potential treatment effects. For whites in the class prime, higher class whites are marginally more likely to think the courts in their area are fair. The difference between low and high class whites here is 4 percentage points (p = 0.066). This is on top of a 5 point class difference in the baseline condition (p < 0.000).
##
## Call:
## lm(formula = court.fair.all ~ court.fair.treat * class.rac, data = cjs.df,
## weights = wts_white)
##
## Weighted Residuals:
## Min 1Q Median 3Q Max
## -1.75490 -0.04959 -0.00782 0.21841 0.78941
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.650394 0.008860 73.406 < 2e-16 ***
## court.fair.treatRace 0.018690 0.012716 1.470 0.141656
## court.fair.treatClass -0.020307 0.012447 -1.632 0.102814
## class.rac 0.054297 0.015789 3.439 0.000587 ***
## court.fair.treatRace:class.rac -0.006945 0.022665 -0.306 0.759291
## court.fair.treatClass:class.rac 0.040979 0.022318 1.836 0.066379 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.2764 on 8084 degrees of freedom
## (3076 observations deleted due to missingness)
## Multiple R-squared: 0.007835, Adjusted R-squared: 0.007221
## F-statistic: 12.77 on 5 and 8084 DF, p-value: 2.188e-12
As for blacks, a different picture emerges. The results below show a sharp divergence in fairness evaluations by class depending on the question wording. For those receiving the class prime, higher class blacks are 9 points less likely to believe the courts in their area are fair than their lower class counterparts (p < 0.05). Interestingly, a similar effect manifests for higher class blacks receiving the race prime, although the magnitude is smaller and imprecisely estimated (\(\beta = -0.07\), p < 0.1).
##
## Call:
## lm(formula = court.fair.all ~ court.fair.treat * class.rac, data = cjs.df,
## weights = wts_black)
##
## Weighted Residuals:
## Min 1Q Median 3Q Max
## -1.2594 -0.1636 0.1036 0.1649 1.2823
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.47710 0.01575 30.286 <2e-16 ***
## court.fair.treatRace -0.00398 0.02242 -0.178 0.8591
## court.fair.treatClass 0.04734 0.02291 2.066 0.0389 *
## class.rac 0.07406 0.02865 2.585 0.0098 **
## court.fair.treatRace:class.rac -0.07068 0.04217 -1.676 0.0939 .
## court.fair.treatClass:class.rac -0.09094 0.04075 -2.232 0.0257 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.3043 on 3066 degrees of freedom
## (8094 observations deleted due to missingness)
## Multiple R-squared: 0.005924, Adjusted R-squared: 0.004303
## F-statistic: 3.654 on 5 and 3066 DF, p-value: 0.002677
We also asked respondents whether give the police more respect would make civilian-police interactions go more smoothly. Higher values denote a belief that being more respectful would lead to more frequent positive interactions. The crosstabs by respondent characteristics suggest that race, not class, shapes these beliefs. Blacks are much less likely than whites to beleif respect leads to consistently positive interactions. 79% of whites believe respect leads to smooth interactions “most of the time” or “always.” In contrast, only 46% of blacks believe this. Consequently, the Chi\(^2\) p-value by race is 0.000. Moreover, within racial groups class does not appear to offer any variation. Perpsectives on this item thus appear to follow more from racial background than class.
## black
## respect.police 0 1
## 0 2 10
## 1 18 44
## 2 49 34
## 3 30 12
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~respect.police + black, d.all), 2) * 100)
## X-squared = 26.657, df = 3, p-value = 6.946e-06
## class.rac
## respect.police 0 0.25 0.5 0.75 1
## 0 6 4 5 3 4
## 1 24 25 26 25 27
## 2 45 46 43 46 46
## 3 25 25 27 26 24
## Warning in chisq.test(round(prop.table(svytable(~respect.police +
## class.rac, : Chi-squared approximation may be incorrect
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~respect.police + class.rac, d.all), 2) * 100)
## X-squared = 1.7315, df = 12, p-value = 0.9997
Whites
## class.rac
## respect.police 0 0.25 0.5 0.75 1
## 0 3 2 2 2 2
## 1 20 20 18 16 17
## 2 48 49 47 51 51
## 3 29 29 33 31 30
## Warning in chisq.test(round(prop.table(svytable(~respect.police + class, :
## Chi-squared approximation may be incorrect
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~respect.police + class, d.wht), 2) * 100)
## X-squared = 1.853, df = 12, p-value = 0.9996
Blacks
## class.rac
## respect.police 0 0.25 0.5 0.75 1
## 0 15 9 11 6 6
## 1 34 39 47 52 53
## 2 37 39 31 32 32
## 3 14 13 11 10 9
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~respect.police + class, d.blk), 2) * 100)
## X-squared = 13.436, df = 12, p-value = 0.3382
Finally, respondents reported whether or not incidents of police corruption were systemic or just “bad apples.” Again, responses vary substantially by race, but not class. 34% of black respondents see these incidents as systemic issues, 23% as bad apples, and 40% a little bit of both. In contrast, 49% of whites focus on bad apples, and only 19% respond that these issues reflect systemic problems. No such variation occurs across class categories. Each class group sees a little over 40% emphasizing bad apples, with between 20 and 26% reponding that it’s a systemic issue. It’s interesting to note that the emphasis on systemic problems rises by class, but the overall distribution doesn’t meaningfully change.
## black
## pol.badapples 0 1
## 1 49 23
## 2 19 34
## 3 30 40
## 4 1 4
## Warning in chisq.test(round(prop.table(svytable(~pol.badapples + black, :
## Chi-squared approximation may be incorrect
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~pol.badapples + black, d.all), 2) * 100)
## X-squared = 16.844, df = 3, p-value = 0.0007608
## class.rac
## pol.badapples 0 0.25 0.5 0.75 1
## 1 42 42 42 43 41
## 2 20 23 23 24 26
## 3 35 33 33 32 31
## 4 3 2 2 1 1
## Warning in chisq.test(round(prop.table(svytable(~pol.badapples +
## class.rac, : Chi-squared approximation may be incorrect
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~pol.badapples + class.rac, d.all), 2) * 100)
## X-squared = 2.6766, df = 12, p-value = 0.9974
Turning to within-group differences, nothing signficantly varies. Even so, there’s interesting descriptive variation within blacks. Higher class blacks are less likely to report that police corruption comes from bad apples, and are more likely to emphasize systemic issues, than are lower class blacks. The proportion reporting that both issues matter stays effectively the same.
Whites
## class.rac
## pol.badapples 0 0.25 0.5 0.75 1
## 1 45 47 49 51 50
## 2 19 20 19 20 22
## 3 34 31 30 28 27
## 4 2 1 1 1 1
## Warning in chisq.test(round(prop.table(svytable(~pol.badapples + class, :
## Chi-squared approximation may be incorrect
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~pol.badapples + class, d.wht), 2) * 100)
## X-squared = 2.7666, df = 12, p-value = 0.997
Blacks
## class.rac
## pol.badapples 0 0.25 0.5 0.75 1
## 1 27 26 22 21 17
## 2 25 29 34 37 41
## 3 42 42 41 41 39
## 4 6 3 3 2 3
## Warning in chisq.test(round(prop.table(svytable(~pol.badapples + class, :
## Chi-squared approximation may be incorrect
##
## Pearson's Chi-squared test
##
## data: round(prop.table(svytable(~pol.badapples + class, d.blk), 2) * 100)
## X-squared = 12.051, df = 12, p-value = 0.4416
Being employed in some level of government helps explain the black-white gap on this outcome as well. First, the marginal effect of employment is larger for blacks than whites. Second, the black-white gap grows smaller for those employed by the government.
##
## Call:
## lm(formula = respect.police ~ employ.gov * black, data = cjs.df,
## weights = wts_whole)
##
## Weighted Residuals:
## Min 1Q Median 3Q Max
## -5.0178 -0.4574 -0.0417 0.5684 3.9277
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 2.048507 0.013215 155.010 < 2e-16 ***
## employ.gov -0.007958 0.033726 -0.236 0.813
## black -0.651980 0.025966 -25.109 < 2e-16 ***
## employ.gov:black 0.258299 0.051652 5.001 5.88e-07 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.7583 on 5793 degrees of freedom
## (5369 observations deleted due to missingness)
## Multiple R-squared: 0.1091, Adjusted R-squared: 0.1087
## F-statistic: 236.5 on 3 and 5793 DF, p-value: < 2.2e-16
A similar pattern holds when looking at variation by whether or not respondents are employed in the criminal justice system. Employment here matters solely for blacks. Employment improves perspectives on respecting the police by half a scale point. The racial gap in evaluations effectively disappears.
##
## Call:
## lm(formula = respect.police ~ employ.cjs * black, data = cjs.df,
## weights = wts_whole)
##
## Weighted Residuals:
## Min 1Q Median 3Q Max
## -5.1531 -0.4157 -0.0395 0.5776 3.8458
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 2.04410 0.01239 165.021 < 2e-16 ***
## employ.cjs 0.05965 0.06096 0.978 0.328
## black -0.61414 0.02280 -26.942 < 2e-16 ***
## employ.cjs:black 0.50946 0.08945 5.696 1.29e-08 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.756 on 5788 degrees of freedom
## (5374 observations deleted due to missingness)
## Multiple R-squared: 0.1141, Adjusted R-squared: 0.1137
## F-statistic: 248.6 on 3 and 5788 DF, p-value: < 2.2e-16
Finally, little systematically varies by a respondent’s specific position in the criminal justice system.
##
## Call:
## lm(formula = respect.police ~ as.factor(cjs.pos) * black, data = cjs.df,
## weights = wts_whole)
##
## Weighted Residuals:
## Min 1Q Median 3Q Max
## -3.04788 -0.46703 -0.05506 0.59481 2.25875
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 2.24026 0.18808 11.911 < 2e-16 ***
## as.factor(cjs.pos)2 0.05701 0.34732 0.164 0.869761
## as.factor(cjs.pos)3 -0.31558 0.37048 -0.852 0.395149
## as.factor(cjs.pos)4 -0.52406 0.27892 -1.879 0.061435 .
## as.factor(cjs.pos)5 0.02360 0.29660 0.080 0.936648
## as.factor(cjs.pos)6 -0.99597 0.26725 -3.727 0.000241 ***
## as.factor(cjs.pos)7 -0.22889 0.32562 -0.703 0.482753
## as.factor(cjs.pos)8 0.17583 0.21679 0.811 0.418107
## black 0.22968 0.25750 0.892 0.373275
## as.factor(cjs.pos)2:black -0.45529 0.44185 -1.030 0.303820
## as.factor(cjs.pos)3:black -0.04564 0.55030 -0.083 0.933966
## as.factor(cjs.pos)4:black 0.20533 0.41006 0.501 0.617002
## as.factor(cjs.pos)5:black -0.02532 0.42479 -0.060 0.952509
## as.factor(cjs.pos)6:black 0.60389 0.38240 1.579 0.115568
## as.factor(cjs.pos)7:black 0.30792 0.47418 0.649 0.516700
## as.factor(cjs.pos)8:black -1.15663 0.30351 -3.811 0.000175 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.8532 on 247 degrees of freedom
## (10903 observations deleted due to missingness)
## Multiple R-squared: 0.2176, Adjusted R-squared: 0.1701
## F-statistic: 4.581 on 15 and 247 DF, p-value: 1.061e-07
Considering variation based on family background, little varies. Blacks on average have less positive views, but nothing varies based on childhood class by either racial group.
##
## Call:
## lm(formula = respect.police ~ chood.class * black, data = cjs.df,
## weights = wts_whole)
##
## Weighted Residuals:
## Min 1Q Median 3Q Max
## -5.0789 -0.4330 -0.0692 0.6145 3.7524
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 2.097891 0.015662 133.947 <2e-16 ***
## chood.class -0.012217 0.009162 -1.333 0.182
## black -0.629802 0.027648 -22.779 <2e-16 ***
## chood.class:black 0.019815 0.016986 1.167 0.243
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.7716 on 11158 degrees of freedom
## (4 observations deleted due to missingness)
## Multiple R-squared: 0.1079, Adjusted R-squared: 0.1077
## F-statistic: 449.8 on 3 and 11158 DF, p-value: < 2.2e-16
Income does more to shape perspectives on respecting the police. Income matters more among whites than blacks, with higher income whites holding more positive views about respecting the police. Income doesn’t matter for blacks. Consequently, with higher income whites becoming increasingly positive, the black-white racial gap increases as income increases.
##
## Call:
## lm(formula = respect.police ~ inc * black, data = cjs.df, weights = wts_whole)
##
## Weighted Residuals:
## Min 1Q Median 3Q Max
## -5.3192 -0.4384 -0.0556 0.6215 3.7927
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 2.001154 0.016013 124.969 < 2e-16 ***
## inc 0.015492 0.002650 5.846 5.18e-09 ***
## black -0.508128 0.027757 -18.306 < 2e-16 ***
## inc:black -0.019256 0.004952 -3.889 0.000101 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.7705 on 11161 degrees of freedom
## (1 observation deleted due to missingness)
## Multiple R-squared: 0.1107, Adjusted R-squared: 0.1104
## F-statistic: 463 on 3 and 11161 DF, p-value: < 2.2e-16
Social Experiences
Police Abused Friends/Family
I being by looking at whether respondents report that they or their peers had been mistreated by the police. Across all items, respondents are less positive in their evaluations of the police. Perhaps more interestingly, across all items the black-white evaluation gap closes as the frequency of mistreatment increases. The gaps remain, but they grow smaller by varying degrees.
Solving Crime
Protecting people like you from violent crime
Treating racial and ethnic groups equally
Not using excessive force on suspects
Holding police officers accountable for misconduct
Summary Evaluation Index
Peers convicted of a Felony
I now turn to conditioning on whether a respondent has friends or family with felony convictions. Across all items, respondents with peers who have experienced a felony are less positive in their evaluations of the police. As with the police mistreatment item, in many cases the black-white evaluation gap closes as the number of peers with convictions increases. The gaps remain, but they grow smaller by varying degrees. Finally, relative to being mistreated by the police, the effect of social connections with felony convictions is smaller.
Solving Crime
Protecting people like you from violent crime
Treating racial and ethnic groups equally
Not using excessive force on suspects
Holding police officers accountable for misconduct
Summary Evaluation Index